Deployment team

Tuesday 24 Jul 2007, 11:00 → 12:00 Europe/Zurich

EVO - GridPP Deployment team meeting

Jeremy Coles

- This is the weekly DTEAM meeting - The intention is to run the meeting in EVO: http://evo.caltech.edu/evoGate/. Join the meeting in the GridPP Community area - The backup plan is to use VRVS - Plane virtual room

Dteam meeting 24/07/07 Attendees: Jeremy Coles Derek Ross Raja Nandakumar Stephen Burke Greig Cowan Barney Garrett Steve Lloyd Frederic Brochu Matt Viljoen Pete Gronbech David Colling Alessandra Forti Yingqin Zheng Experiment problems/issues ========================== LHCb RN: Testing SL4 CE at Tier 1, Clearing some problems with SAM tests. Atlas SL: working out problems with 13.10 Follow up RB issues with Graeme SB: Castor testing is going, testing by Brian Davies + CMS, in reasonable shape, starting general Atlas testing from CERN. Disk storage back in dCache, but production has not been restarted. Cosmic run Aug-Sep, are Atlas expecting more transfers to T2 in this period? Depends, may do transfers to some T2 but probably not UK ones CMS DC: Stuart going ahead with Production, problems with QMUL Castor testing going reasonably well Site Readiness Review actions ============================= Concern with 2hr response time ------------------------------ Site concerned they can meet this response time, not clear if its problem closed or just responded to. RSS feeds --------- Will T2 admins use RSS feeds? How many do already? What information should be made available Sites experience with GGUS -------------------------- How fast are sites responding to GGUS? Are sites still unhappy with GGUS? London sites have lots of unclosed tickets for fixed problems. Possibly Olivier was doing these. Sometimes sites are waiting for users to confirm problem fixed, may close tickets after a period of time to avoid this. Users can't close or reopen tickets in GGUS. How to improve Prod-PPS team links ---------------------------------- Do we want a summary of issues in the PPS, do we want them to report new functionality , new config files Lack of people testing things on PPS, few people running jobs - this is a higher level issue ROC Update ---------- TPM shifts - Team 7 13 Aug and Team 8 20 Aug COD - Backup 30 Aug and Lead 3 Sep RGMA Registry ip address change - UK Impacts? No outbound firewall rules in London, Southgrid - Birmingham iptables. Tier 1 will probably give 2 weeks notice in September Priority voting ---------------- Currently only have 3 votes assigned, now have 5, need feedback from sites, will raise at TB-Support meeting. Push for new VOs ---------------- SL4 migration may be a good time to add new VOs NGS-GridPP interoperation ========================= Glasgow and Imperial are now affiliates to NGS. what are the requirements to become affiliated? Jeremy is arranging a meeting to discuss SLAs with NGS NGS vo (ngs.ac.uk) should be enabled at sites, it replaces the old ngs vo, which allows all users of the NGS to submit jobs to gridpp, but it may be useful to have a separate vo to allow testing of interoperation, possibly the GIN vo could be used. Oxford have ngs.ac.uk enabled, investigating as to whether stagein is required for pbs/torque. This will break when we move to gLite CEs MPI experiences: Grid Ireland are interested. Glasgow have implemented it, Imperial are looking at it. Is anyone using MPI_START - this is what is used on NGS. Policies - Don't have clear policies on who signs agreements that the VO accepts the AUP of the NGS. More sites should enable ngs.ac.uk How does this vo get used to test the site? AOB === What is Atlas status in testing T1 to T2 links? Investigating why transfers to UK are suspended, but no answer from CERN yet. Usual production streams will flow from T2 to T1. VO Queue length - Assuming the 24h CPU and 36h WCT are normalised to 1 KSPECINT2k, but need to confirm this. Raja to follow up in next meeting if he uncovers any more info on how LHCb specifies job length Stalled jobs policy now at v1.0, will be applied from 1 Aug to end of year, guideline for sites - not mandatory to kill jobs, but procedure should be followed if they are. QR status - all received but London UKI Middleware status - revisit on Thursday, need to track how sites are keeping up with updates Oxford biomed job problems - check config, set max number of jobs queued.It seems Biomed target sites, will see if other sites have similar problems at T-Support meeting

11:00 → 11:20 Experiment problems/issues

Review of weekly issues by experiment/VO - LHCb - CMS - ATLAS - Other

11:20 → 11:35 Site Readiness Review actions

Some things to be followed up - Concerns with 2-hour response time - Using RSS feeds to distribute information - Review of site experiences wrt GGUS - How to improve the Production-PPS team links

11:35 → 11:40 ROC update

- TPM https://gus.fzk.de/pages/tpm.php. Team 7: 13/08 and Team 8: 20/8 - COD https://cic.gridops.org/index.php?section=cod&page=homepage. Backup site 30/7. Lead site 3/09. - Changeover of RGMA registry hardware - new ip address. What are the UK impacts? This is mainly an issue for sites with firewall rules on outbound connections. - ROC priorities voting: https://twiki.cern.ch/twiki/bin/view/EGEE/SA1_TCG - please check item 10 as there has been feedback - Push on new VOs (e.g. supernemo.vo.eu-egee.org) - can T2Cs take responsibiity for managing tickets in their region. I can set up regular footprints reports if this would help, it works very well with RAL

11:40 → 11:55 NGS-GridPP interoperation

- Review of activities along with problems and issues - Recording information in the wiki http://www.gridpp.ac.uk/wiki/Interoperation_activities

11:55 → 12:00 AOB

- Experiment transfers will flex network connections => Andrew to work closely with Fredric and James/Simon - VO queue lengths - what is acceptable/most useful. Should we have a UK default? ..... "After discussion with ..., we decided to ask for shortening *GRID* ATLAS queues at CERN. New values are 24h CPU and 36h WCT". Proposal is to ask for site views on Thursday. - Stalled jobs policy goes to version 1.0. To be discussed at Thursday's UKI meeting. This will be the policy for 1st Aug - 31st December in the first instance. - Quarterly reports status - Question on UKI middleware status: gLite-3_1_0 (2 sites); gLite-3_0_2 (5 sites); gLite-3_0_1 (1 site) and gLite-3_0_0 (16 sites) - Oxford biomed job problems (900 waiting jobs: 20 slots). Batch scheduler configuration proposal.

document GridPP-PMB-113-Inefficient_Jobs_v1.0.doc GridPP-PMB-113-Inefficient_Jobs_v1.0.pdf